DeepMind Unveiled Scientific Tools for Gemini: Hypothesis Generation and Experiment Automation

DeepMind unveiled a suite of experimental tools embedded in Gemini to accelerate scientific research. The announcement was made at Google I/O 2026. The three prototypes are designed to scale and improve the precision of experiments in biology, climatology, and other fields.

Three Tools for Science

Hypothesis Generation operates on the principle of an "idea tournament"—analyzing millions of scientific papers and generating competing hypotheses, selecting the most promising ones. The system verifies each conclusion against sources, reducing the risk of errors and ensuring result verifiability. This is particularly useful in fields where the volume of literature grows faster than a person can read.

Computational Discovery is based on the AlphaEvolve and ERA systems—automatically creating and testing thousands of code variants simultaneously. This accelerates complex computational experiments in molecular dynamics, optimization, and machine learning, which typically require hours of manual work. The system finds not only working solutions but also non-obvious optimizations.

Literature Insights integrates NotebookLM to structure scientific literature. The tool transforms stacks of papers into tables, summaries, and visualizations—for quickly identifying patterns and gaps in knowledge.

Where It's Already Being Used

Companies BASF and Klarna use AlphaEvolve to optimize supply chains and machine learning models. Research institutions are applying these tools to investigate rare genetic diseases, antibiotic resistance, and other complex biological questions.

• Rare genetic diseases—analysis of protein properties through AlphaFold
• Climatology—solar energy potential forecasting
• Bioengineering—search for plastic-degrading enzymes for waste utilization
• Epidemiology—disease spread modeling and intervention assessment

Science Skills: A Complete Data Ecosystem

The new Science Skills module integrates over 30 scientific databases: AlphaFold, UniProt, PubChem, and others. Previously, researchers had to manually load data and insert it into context. Now everything is built in—Gemini searches the correct sources and returns verified results. This reduces analysis time from hours to minutes and guarantees citation accuracy.

What This Means

The boundary between information retrieval and interpretation is blurring. Gemini for Science handles data preparation, literature structuring, and hypothesis selection—freeing researchers for the creative aspects. The tools are currently experimental, but within a year or two, these capabilities will become standard in research organizations and corporate labs.